Chapter 4
Downloaded by UNIV OF MINNESOTA on October 14, 2014 | http://pubs.acs.org Publication Date (Web): June 3, 2013 | doi: 10.1021/bk-2013-1133.ch004
Characterization of CO2 Behavior on Rutile TiO2 (110) Surface Yeohoon Yoon* Fundamental and Computational Science Directorate, Pacific Northwest National Laboratory, Richland, Washington 99352 *E-mail:
[email protected] The dynamic behavior of carbon dioxide (CO2) adsorbed on the rutile TiO2 (110) surface is studied by dispersion corrected density functional theory (DFT) and combined ab initio molecular dynamics (AIMD) simulation. Understanding the behavior of CO2 is important regarding possible applications for treating CO2 in current environmental problems along with the consideration as a renewable energy source. Concerning the ability as a reducible support of TiO2 surface, a fundamental understanding of the interaction between CO2 and TiO2 surface will help extending the possible applications. In the current study, CO2 interaction and dynamic behavior on the TiO2 surface is characterized including the effect of the oxygen vacancy (OV) defect. Also the coverage dependence of CO2 behavior is investigated since more contribution of the intermolecular interaction among CO2 molecules can be expected as the coverage increasing.
Introduction The increasing energy demand requires the development of new energy sources and it is expected to be environmentally sustainable at the same time (1). Hydrocarbon fuel is currently a primary source of energy since it is provided by nature with its ease of transportation and storage. Unfortunately, the amount of this resource is limited and the combustion of these fuels carries the significant © 2013 American Chemical Society In Applications of Molecular Modeling to Challenges in Clean Energy; Fitzgerald, G., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2013.
Downloaded by UNIV OF MINNESOTA on October 14, 2014 | http://pubs.acs.org Publication Date (Web): June 3, 2013 | doi: 10.1021/bk-2013-1133.ch004
environmental pollution including carbon dioxide (CO2). Concerning a particular impact of CO2 to the global warming through the greenhouse effect, there is an increasing interest in technological solution to alleviate CO2 emission (2, 3). Some of such strategies concern the possibility of capture, storage and sequestration. Additionally, a possibility of converting CO2 into a valuable fuel should be an important application among CO2 treatments. For facilitating such solutions, a fundamental understanding of CO2 activation is required while this activation should overcome the extremely strong chemical bonds accompanied by the carbon atom. Considering a possible role of surface including oxide surface as a (photo-) catalyst and a support of reducibility, only little is known about the interaction of CO2 with the surface (4). Accomplishing such understanding of fundamental interaction requires the ability of design and modeling at the level of atoms. This kind of detailed knowledge can be provided by the theory and computation so that such molecular modeling can be applied to challenges in clean energy problems (5). While the collective observables are obtained commonly in the laboratory, the computational modeling allows us to examine in more detailed manner at the atomic level. In regards to the physical origin of CO2 activation and related interactions, the contributions by electronic and thermal environment are necessary to be investigated. Contemporary quantum mechanical approach based on the density functional theory (DFT) is able to address such points and combining molecular dynamics technique can provide macroscopic ensemble nature which can be compared directly to the observation in the laboratory (6, 7). For the applications by taking advantage of oxide surfaces, a specific interest can be focused on TiO2 surface with the motivation initially arisen from the capability of its photo reduction of CO2 to hydrocarbons or methanol (3, 8, 9). Besides, CO2 has been utilized for characterizing different sites on TiO2 surface (10) as well. To date, rutile TiO2 (110) surface is the most studied single crystal surface due to its stability (11), and the characterization of CO2 on rutile TiO2 (110) surface has been studied extensively so far (10–16). Concerning the adsorption site on the surface, CO2 initially adsorbs weakly on oxygen vacancies (OV) and subsequently populates on five-coordinated Ti (Ti5C) sites (11). The CO2 molecule adsorbed on OV can be reduced to CO by applying energy above ~1.8 V which can be observed, for example, in the scanning tunneling microscopy (STM) experiment with the injected electrons from tip (13, 14, 16). Alternatively, it can also diffuse from OV with barrier energy of 0.14 eV (12). Furthermore, the CO2 on Ti5C site is shown to be mobile with relatively low diffusion barriers (12). In this chapter, the basic understanding of CO2 adsorption is addressed quantitatively by means of DFT and combined AIMD simulation. Then a detailed character of the coverage dependent binding configurations of CO2 and their dynamic behaviors are investigated. The organization of this chapter is as follows: The theoretical method and the model system are explained in Method section along with the verification for their suitability. In the Results and Discussion section, the single CO2 adsorption and diffusion on TiO2 surface is addressed firstly, and then the effect of OV is discussed followed by the characterization of CO2 according to the different coverages. In the Conclusions section, all results are summarized and overall collective perspective is discussed. 52 In Applications of Molecular Modeling to Challenges in Clean Energy; Fitzgerald, G., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2013.
Methods
Downloaded by UNIV OF MINNESOTA on October 14, 2014 | http://pubs.acs.org Publication Date (Web): June 3, 2013 | doi: 10.1021/bk-2013-1133.ch004
Model System The TiO2 surface is flexible to its electronic environment so that it becomes acidic if there are excess electrons, and that happens commonly when studying materials as a reducible support concerned in the energy related research (2, 17). When modeling such system, cares should be taken to describe proper electronic environment at the surface. The current model system is constructed in order to avoid large dipole moment in the slab and provide reasonable work function within the adopted theoretical method. Such requirement can be achieved by using larger number of layers and six TiO2 tri-layers are applied in the current study. In addition, the bottom Ti layer is fixed to their bulk lattice positions while other layers are allowed to relax. The validity is confirmed with consistent electrostatic (Hartree) potential along the direction of surface normal. Also a 10 Å thick vacuum layer is applied above the surface to minimize electrostatic interaction between periodic images in the same direction. For the surface area included in the model, (3×4) rutile TiO2(110) slab model is considered as a surface of wide enough and this model has been focused on with its stability. The large area of the model surface makes it possible to avoid the intermolecular interaction between single isolated CO2 molecule and its periodic image, while many CO2 molecules in high coverage can be described regardless of this factor by nature of periodic boundary condition. For the modeling of various coverages, four different coverages are considered by placing 4, 6, 8, 12 CO2 molecules on the slab which represent 1/3, 1/2, 2/3, 1 monolayer (ML), respectively. Particularly when a possible oxygen vacancy (Ov) defect is considered, one Ov is considered within the size of the current model, in which the concentration of OV corresponds to 0.083 ML, and it is reasonable to be compared with the experimental observation (18, 19). Computational Details For the electronic structure calculation, density functional theory (DFT) is adopted as implemented in the CP2K package (20–22) with gradient corrected (PBE) functional for exchange and correlation (23). Norm-conserving pseudopotentials are used to describe core electrons (24). The wave functions are constructed for the calculation of electrostatic energy by expanding a double-ζ Gaussian basis set in which basis superposition errors are minimized (25) combined by an additional auxiliary plane wave basis with 400 Ry energy cutoff. Brillouin zone sampling is performed by using Γ-point. In order to describe more precisely for the long range interaction among adsorbed molecules and the surface, the dispersion forces are modeled by the DFT-D3 method (26). By doing this, an attractive dispersion coefficient and the corresponding cutoff function are added to the DFT potential using the standard parameter set by Grimme (26) with a cutoff of 10 Å. The effect of dispersion correction is observed as up to 0.05 eV increase of the binding energy difference among various CO2 configurations and this approach is believed to represent adequate potential energy surfaces for hydrocarbons on oxides (27, 28) as well as the structural and spectroscopic properties of liquid supercritical CO2 based on the previous studies 53 In Applications of Molecular Modeling to Challenges in Clean Energy; Fitzgerald, G., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2013.
Downloaded by UNIV OF MINNESOTA on October 14, 2014 | http://pubs.acs.org Publication Date (Web): June 3, 2013 | doi: 10.1021/bk-2013-1133.ch004
(29, 30). When considering kinetic coordinates, the climbing image nudged elastic band method (CI-NEB) (31) is adopted employing 13-17 replica. For a finite temperature sampling, ab initio molecular dynamics (AIMD) simulations are performed within a canonical (NVT) ensemble using Nosé-Hoover thermostat (32, 33) with a time step of 0.5 fs during more than 20 ps of equilibrated trajectory at the temperature of 130 K. Note that this temperature is near the desorption temperature so as to increase the ability of sampling relevant configurational space within relatively short time (20 ps) duration of trajectory due to the limitation of AIMD. By doing so, we are able to observe CO2 rotation and diffusion events from the sampled trajectories, yet not the desorption events. When thermally equilibrated configurations are quenched, the simulated annealing technique is adopted by rescaling the velocity with a factor of 0.99 at each time step. For the calculation of the electrostatic potential at each time step, DFT calculation is performed as explained above. For the model of a reduced TiO2 surface, an accurate description is required for the excess electrons produced by OV. For this case, DFT+U method (34) is used with an effective U parameter (Ueff) applied to the Ti 3d electrons within a local spin density approximation for the spin polarized formalism. While focusing on the correct description of the work function and relative band gap position, test calculations have been performed as shown in Figure 1. When large value is chosen as effective U parameter, the location of defect state is fairly well separated from the conduction band so that excess electrons are easily localized at that state. On the other hand, the work function is increasing by increasing effective U parameter, which results in rather unreal nature of the role of excess electrons. Therefore an optimal condition is required to be defined first, and in the current study, the effective U parameter of 4.1 eV is adopted since an adequate work function (4.9 eV) (35) was able to be reproduced while the location of defect states is still reasonably described based on the expected position at 0.9 eV below conduction band (36, 37).
Figure 1. Electronic structures depending on the different Ueff values within DFT+U method. The energies are relative to the energy at vacuum state and the work functions are obtained by the difference between the vacuum and defect state energy. These electronic structures are obtained from the (3×4) TiO2 (110) slab model with six tri-layers and one OV defect on the surface. 54 In Applications of Molecular Modeling to Challenges in Clean Energy; Fitzgerald, G., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2013.
Results and Discussion
Downloaded by UNIV OF MINNESOTA on October 14, 2014 | http://pubs.acs.org Publication Date (Web): June 3, 2013 | doi: 10.1021/bk-2013-1133.ch004
Single CO2 When a single CO2 molecule is adsorbed on the TiO2 (110) surface, it prefers to bind on OV site and heals this vacancy defect first. Then additional CO2 molecules begin to populate on the five-coordinated Ti sites (Ti5C) as the coverage increased (13–16). Therefore, when considering single CO2 adsorption on the surface in the present study, OV sites are assumed to be healed so that CO2 molecule on the Ti5C row on the clean surface is able to be considered. As a first assessment of the behavior of CO2, numbers of single CO2 binding configurations to the TiO2 surface are investigated. Figure 2 shows several noteworthy configurations including the energetically most stable configuration as shown in Figure 2(a). In this configuration, CO2 binds with one O atom to top of a Ti5C site while the O=C=O molecular axis is tilted by a polar angle, ψ'45° with respect to the surface normal direction and by an azimuthal angle, φ=90° along the [11̅0] direction toward neighboring bridge bond oxygen (Ob) at the same time. The binding energy can be calculated by subtracting the energy of gas phase CO2 and bare TiO2 slab from the energy of whole CO2 adsorbed TiO2 slab model, and the resulting binding energy of the most stable configuration is obtained as 0.45 eV. It is in good agreement with previous DFT calculation in which it was 0.44 eV or 0.39 eV depending on the different methods of dispersion correction (17). The other configurations in Figure 2 are considerable in the sense of possible diffusion path of adsorbed CO2 molecule. First of all, the rotation of CO2 is regarded as one of the possible dynamic motion. In this case, dangling O atom out of the surface (denoted Od) rotates while keeping the other bound O atom to top of a Ti5C site as an anchor (denoted Oa). In Figure 2(b), CO2 molecule has rotated azimuthally by φ≈45° with respect to the [11̅0] direction and the energy difference to the most stable one is just 0.02 eV which can be overcome easily in thermal condition. In other words, the initiation of the rotation of adsorbed CO2 about Ti5C-Oa axis is thermally facile. Note that the Od atom tries to come close to the neighboring Ti lattice during rotation, and as a result, the length of O=C=O projected to the surface normal, i.e. molecular height should change as well. During the rotation, CO2 molecule is supposed to be able to pass through the configuration parallel to the surface with ψ=90°, φ=0° in which each O atom binds at the distance of 2.33 Å to two neighboring Ti5C sites (Figure 2(c)). In this configuration, the molecular height should be close to 0 Å. In order to reach this configuration from the most stable one, it is required the energy increasing by 0.03 eV. When CO2 molecule happens to have this configuration during the thermal motion, one can expect the possibility of its tumbling to the neighboring Ti5C site by exchanging the role of O atom as an anchor. Besides, when CO2 molecule is adsorbed to Ob site as shown in Figure 2(d), the binding energy is 0.24 eV smaller than that of the most stable configuration. Thus Ob site can be excluded from the concerns of the most probable single CO2 adsorption site.
55 In Applications of Molecular Modeling to Challenges in Clean Energy; Fitzgerald, G., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2013.
Downloaded by UNIV OF MINNESOTA on October 14, 2014 | http://pubs.acs.org Publication Date (Web): June 3, 2013 | doi: 10.1021/bk-2013-1133.ch004
Figure 2. Single CO2 molecule adsorption configurations on TiO2(110) surface. Four representative configurations are shown and their binding energies are (a) 0.45 eV, (b) 0.42 eV, (c) 0.33 eV and (d) 0.21 eV. (For interpretation by color, see the web version of this article.)
Regarding the facile thermal rotation and the probable tumbling motion along the Ti5C row as a hypothetical diffusion mechanism, the kinetic barrier energies of such kind of motions are examined by CI-NEB as shown in Figure 3. When CO2 starts rotating from the most stable configuration, it can reach to the intermediate configuration corresponds to Figure 2(c). From this configuration, the molecule can reverse the process or rotate further to end up back to the most stable configuration. This rotation motion requires 0.05 eV activation energy between two symmetrically equivalent stable configurations which is represented by a rotation about the Ti5C-Oa axis by Δφ=180°. Alternatively, it may be able to tumble to the neighboring Ti5C site through this intermediate configuration. In this case, the molecule switches its anchoring Oa atom to Od atom while Od atom becomes anchor Oa bound to top of the neighboring Ti5C site, then it keeps rotating about new generated Ti5C-Oa axis. This process exhibit an energy barrier of 0.06 eV and this implies extremely fast diffusion even at low temperatures. In fact, this diffusion along the Ti5C row is competitive with the rotation mechanism described earlier in terms of the activation energy. The remaining possible diffusion path on the surface should be related to the path involving Ob sites. The diffusion path across the Ob row shows the barrier energy of 0.21 eV and it is much higher than others considered previously. Thus it can be disregarded from the probable dominant diffusion process on the surface while the diffusion can be conclusively characterized by the combined rotation and tumbling motion along the Ti5C row. 56 In Applications of Molecular Modeling to Challenges in Clean Energy; Fitzgerald, G., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2013.
Downloaded by UNIV OF MINNESOTA on October 14, 2014 | http://pubs.acs.org Publication Date (Web): June 3, 2013 | doi: 10.1021/bk-2013-1133.ch004
Figure 3. Diffusion energy barriers corresponding to three different diffusion pathways and their schematic cartoon with (a) rotation and tumbling, (b) pure rotation and (c) crossing Ob row. (For interpretation by color, see the web version of this article.) Effect of OV Defect on Single CO2 Binding In the previous section, a single CO2 binding to the TiO2 surface is considered assuming all possible OV defects are already healed while a CO2 molecule is fairly far away from such OV. However, there could be some differences if a CO2 molecule happens to adsorb on the OV defect site or reside near that site in the presence of the CO2 molecule already adsorbed on it. When considering the defected surface, there are two excess electrons generated by OV defect. In order to account for proper charge distribution by those electrons (38), DFT+U method (34) is applied to determine the configuration and the binding energy. As a first result, the binding energy of CO2 on OV site is found to be 0.60 eV and this shows that the binding is 0.17 eV stronger than the strongest binding to regular Ti5C site. It supports the assumption that a CO2 is preferentially adsorbed on OV defect site first. Then the CO2 molecule adsorbed next to another CO2 already bound to an OV site is considered as shown in Figure 4. If they are placed two Ti lattice distance (2×2.96 Å) away as a model of fairly isolated molecule (Figure 4(a)), the binding 57 In Applications of Molecular Modeling to Challenges in Clean Energy; Fitzgerald, G., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2013.
Downloaded by UNIV OF MINNESOTA on October 14, 2014 | http://pubs.acs.org Publication Date (Web): June 3, 2013 | doi: 10.1021/bk-2013-1133.ch004
energy is 0.43 eV which is similar to the binding energy of single CO2 on a regular Ti5C site. Note that this binding energy is 0.02 eV lower than on clean surface though, as a result of the DFT+U description of TiO2 slab which is opposed to the normal DFT description. However, the binding energy increases to be 0.48 and 0.49 eV as CO2 coming close to another CO2 on OV site respectively to the different configurations (Figure 4(b) and (c)). Thus it is confirmed that there is stronger binding of CO2 next to OV adsorbed CO2 about 0.05 and 0.06 eV comparing to the binding on the regular Ti5C site. When considering the reason of such stronger binding, it can be interpreted that the intermolecular interaction between CO2 molecules is dominant while the electrostatic energy from the excess electron is almost disappeared on the surface region by healing OV. The resulting two most stable binding configurations also support this when compared to two typical liquid CO2-CO2 interactions through T-shaped and slipped parallel configurations (29, 39). In summary, one can see stronger binding of CO2 on the Ti5C site near another CO2 adsorbed on OV but the binding energy is still enough for it to make a thermal diffusion such as the one described in the previous section.
Figure 4. Single CO2 adsorption configurations on TiO2 surface in the presence of neighboring CO2 adsorbed on OV and their binding energies are (a) 0.43 eV, (b) 0.48 eV and (c) 0.49 eV. (For interpretation by color, see the web version of this article.) Coverage Dependence of CO2 Binding Configurations Although the single CO2 interaction with TiO2 surface and its dynamic behavior are intensively investigated, things shall be changed if there are more CO2 molecules on the surface which can interact to one another. In order to understand such an adlayer configurations at higher coverages, model systems with four different coverages (1/3, 1/2, 2/3, and 1 ML) are constructed and the ensembles of those configurations are sampled by AIMD simulations at the finite temperature of 130 K. In general, the Ti5C bound CO2 are free of motion including on-site rotation and diffusion at low coverage while their configurations become more restricted at high coverages. Even though the diffusion rate is hardly quantified due to the limit of statistics from AIMD simulations, several CO2 hopping events are observed from one Ti5C site to the next during 20 ps sampling. In the lowest coverage, i.e. 1/3 ML case, 0.125 counts/ps hopping events are observed per CO2 molecule 58 In Applications of Molecular Modeling to Challenges in Clean Energy; Fitzgerald, G., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2013.
Downloaded by UNIV OF MINNESOTA on October 14, 2014 | http://pubs.acs.org Publication Date (Web): June 3, 2013 | doi: 10.1021/bk-2013-1133.ch004
which is almost 10 times more than the one from 1 ML trajectory. Assuming Arrhenius formula is applicable with the preexponential factor of 1×10-13 s-1, the hopping energy barrier of 1/3 ML coverage is 0.055 eV which is very close to the value of 0.050 eV obtained by CI-NEB calculation as shown in Figure 2. This energy barrier is enough to activate the diffusion motion from the energetically most stable configuration (Figure 2(a)) in thermal condition around at 130 K. When looking into overall features of sampled ensemble configurations at all coverages, basically no well-ordered ones are observed from AIMD, indicating the structures are characterized by a component of dynamic disorder. A representative set of configurations observed at different coverages are considered in order to investigate the relation between such disordered and ordered configurations. For the ordered configuration, CO2 molecules are placed to be hypothetically well ordered based on the single CO2 adsorption configuration. And then this is thermalized at the temperature of 50 K followed by a slow simulated annealing to the temperature of 0 K. By doing so, well-ordered thermal configuration can be quenched while avoiding major rotation and translational motion. On the other hand, one random configuration from the AIMD trajectory is extracted and quenched to temperature of 0 K as a representative set of disordered configuration. During this simulated annealing, one can expect the maximal structure optimization while suppressing possible energy interconversions seen at higher temperature and the resulting configurations are shown in Figure 5. And then the energy differences of ordered and disordered configurations are assessed according to the different coverages. When comparing the binding energy per molecule based on the configurations obtained as above, ordered configurations are generally less stable by 0.06 to 0.09 eV/CO2 than the disordered dynamic configurations from the equilibrated ensemble. In principle, such energy difference is small within the kinetic energy distribution at this temperature so that both of configurations are energetically feasible. However, disordered configurations become more populated concerning free energy nature accompanying entropic effect. As a result, only partially ordered configurations are observed during the simulation. If one look into more in detail of the configurations from Figure 5, the popular configuration is the CO2 bound to Ti5C site while being toward Ob direction with tilting in the possible range of azimuthal rotation angle. Even though the majority of configurations are tilted CO2, note that there is a non-negligible fraction of CO2 lying in between two neighboring Ti5C sites (φ≈0°) and this configuration is able to stabilize the neighboring CO2 next to it. We have discussed observed CO2 configurations so far, as a single molecule separately. Now the averaged feature from the ensemble of configurations will be considered, which are observed practically in most cases of experiments. As a meaningful quantity describing ensemble configurations, the distributions of CO2 orientation is considered as shown in Figure 6. This distribution can be obtained through the 3-dimensional maps of projected Od positions relative to Ti5C bound Oa position onto the TiO2(110) surface plane. The CO2 molecules which have no O atom within 2.5 Å from any Ti5C site are excluded from the distribution. The distance criteria of 2.5 Å is obtained from the first nearest neighbor distance of radial distribution of Ti5C-Oa distance. 59 In Applications of Molecular Modeling to Challenges in Clean Energy; Fitzgerald, G., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2013.
Downloaded by UNIV OF MINNESOTA on October 14, 2014 | http://pubs.acs.org Publication Date (Web): June 3, 2013 | doi: 10.1021/bk-2013-1133.ch004
Figure 5. The representative snapshots of “disordered” configurations extracted from the 130 K AIMD trajectories followed by being quenched to 0 K. The “ordered” configurations constructed based on the most stable single CO2 at the temperature of 50 K at four different coverages. (For interpretation by color, see the web version of this article.)
In general, there are two mainly populated orientations at φ=0° and in the range of φ=45~90° with maximum near φ≈65° for all coverages. The former configuration can be interpreted as the one laying down flat, i.e. ψ=0°, on the surface. This type of configuration can be easily found during the AIMD simulation as shown in the snapshot from Figure 5. The coverage dependence of the population of this flat lying configuration is decreasing from about ~25% at 1/2, 2/3 ML to ~10% at 1 ML which indicates the contribution of CO2-CO2 interaction is decreasing as the coverage is increasing. The distance of this configuration from the surface is approximately the same as the height of C atom of tilted standing CO2 in φ=45~90° range so that it can generate T-shaped dimer which is a common feature in both liquid and solid CO2 structures (29, 40). The second most populated configuration is distributed in rather wider space range while it represents to be oriented φ=45~90° with polar tilting ψ≈45° with respect to surface normal. This configuration is similar to most stable single CO2 configuration (Figure 1(a)). The rotation between two configurations is feasible with a barrier energy (ΔE) of 0.06 to 0.07 eV at all coverages based on , where Pi is the population the population ratio by of configuration i. Note that this barrier energy is similar but a little higher than that of single CO2 rotation possibly due to the contribution of the intermolecular CO2-CO2 interaction. 60 In Applications of Molecular Modeling to Challenges in Clean Energy; Fitzgerald, G., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2013.
Downloaded by UNIV OF MINNESOTA on October 14, 2014 | http://pubs.acs.org Publication Date (Web): June 3, 2013 | doi: 10.1021/bk-2013-1133.ch004
Figure 6. The distributions of projected position of Od relative to Oa onto the TiO2(110) surface plane for CO2 molecules that are bound to Ti5C sampled from AIMD simulations at 130 K. The x- and y-axis represent the surface plane with [01̅0] and [01̅0] directions and z-axis represents the population of Od at that position. The resulting arc shape distributions correspond to the distribution by the azimuthal angle φ. (For interpretation by color, see the web version of this article.)
There is one more noteworthy feature from Figure 6 particularly at 1 ML coverage exhibiting an additional narrow band in the range of φ=60~90° at longer Oa-Od distance around 2.3 Å while the population coming from this configuration is non-negligible with around 9%. This type of configuration can be assigned as CO2 molecules on the Ob row in parallel to the surface plane as it can also be found from the snapshot from Figure 5(d). If this configuration is taken out as a single molecule, the binding energy in the temperature of 0 K can be calculated to be 0.23 eV higher than that of the most stable single molecule configuration as shown in Figure 1(a) and (d). However, when the binding energy per molecule at 1 ML is considered after being quenched to the temperature of 0 K, it becomes larger to be 0.45~0.49 eV as comparable to the strongest binding energy of single CO2 on regular Ti5C site (Figure 1(a)). Therefore, two bound configurations either at Ob or Ti5C can be in thermal equilibration and this implies that the interaction between CO2 molecules participates in stabilizing the configuration on Ob at 1 ML. In order to examine relative positions of neighboring CO2 molecules beyond overall ensemble configuration, their relative spacing is examined as well by pair distribution functions (g(r)) while all g(r) are normalized with respect to the 61 In Applications of Molecular Modeling to Challenges in Clean Energy; Fitzgerald, G., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2013.
Downloaded by UNIV OF MINNESOTA on October 14, 2014 | http://pubs.acs.org Publication Date (Web): June 3, 2013 | doi: 10.1021/bk-2013-1133.ch004
volume of a hemisphere. When O atoms are within the first nearest neighbor distance of Ti5C-Oa distance distribution (2.5 Å), they are considered as anchor Oa whereas the others should be dangling Od. Then the pair distribution function of Oa-Oa distance is considered first among all CO2 molecules on the surface (Figure 7(a)), and this quantity shows the average spacing among CO2 molecules. For the Od distribution, the population along the Od-Od distance is counted within the first nearest neighboring Od-Od pair distribution peak (Figure 7(b)) and this represents the distribution of relative tilting direction of one another. In the case of low coverages such as 1/3 and 1/2 ML, Oa-Oa distribution shows that two neighboring CO2 molecules reside on next to each other approximately 63 and 75 % of the time, respectively. This implies that CO2 intermolecular interactions are considerably large even when they are spatially well distributed initially with the binding to surface Ti5C site. Even though there is a free space for CO2 can visit, they prefer to gather to have a dimer-like configuration due to the CO2-CO2 interaction. Regarding Od-Od population distribution, a broad feature is shown between one to two Ti5C lattice spacing which corresponds to 2.96~5.92 Å with maximal population near 5 Å. This implies that two neighboring CO2 molecules prefer to tilt away from each other. At the high coverage, Oa-Oa distribution becomes rigid with one Ti5C lattice distance as expected and Od-Od distribution shows sharper peak near 4.5~5.0 Å which is almost the same as the one in lower coverages. The sharper peak indicates the geometric structure is more rigid and in other words, the structure of overall CO2 molecules becomes crystal-like with each CO2 azimuthally tilted.
Figure 7. (a) g(r) of Oa in the range corresponding to first nearest neighbor distances. The relative populations by integration are 1:1.27:1.28:1.59 for the coverages of 1/3, 1/2, 2/3 and 1 ML, respectively. (b) The population of the Od -Od distance from their corresponding Oa within the range of the first peak from (a). 62 In Applications of Molecular Modeling to Challenges in Clean Energy; Fitzgerald, G., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2013.
Downloaded by UNIV OF MINNESOTA on October 14, 2014 | http://pubs.acs.org Publication Date (Web): June 3, 2013 | doi: 10.1021/bk-2013-1133.ch004
We have investigated so far the dynamic nature of adlayer configurations due to non-negligible interaction among CO2 molecules comparable to that between CO2 and TiO2 surface. Indeed, the quantification of such interactions will provide the physical origin of the interactions. For that reason, the energetics of those two interactions are considered by pair distribution functions (g(r)) again for the relevant pairs of interactions. For the interactions among CO2 molecules, a pair of C of CO2 and O of another CO2 (OC) is taken into account, while another pair of C of CO2 and bridge bond O from TiO2 surface (Ob) is considered for the interaction between CO2 and TiO2 surface. When considering the first nearest neighbor peak, both types of contact shows at the similar location around 3.0 Å regardless of the coverage and it implies that those two interactions are competitive on the TiO2 surface. As shown in Figure 8, these pair distribution functions can be converted to potential of mean force (PMF) by A(r) = −kBT ln(g(r)) which allows us to quantify the magnitude of the interactions between these species. Here kB is the Boltzmann constant and T is the temperature (41). Particularly, 1 ML configuration is focused on since it is expected to be the most ordered and possible maximum interactions should be contained at this coverage. As a result, the binding free energies of CO2-CO2 and CO2-TiO2 surface are 0.05 eV and 0.06 eV, respectively. Although the strength of those interactions are smaller than typical Ti5C-Oa interaction (ca 0.40 eV) inferred from binding energies, they still should be considered as non-negligible magnitude particularly when they contribute collectively. Comparison of these interactions to the ones in gas phase will provide the physical character of interactions while gas phase interactions are well classified (30). There are two typical interactions in gas phase; one is T-shaped quadrupole-quadrupole interaction and the other is slipped parallel dipole-dipole interaction. Those interactions are able to be quantified with binding free energies of 0.06 eV and 0.07 eV respectively, and they are quite similar to currently obtained CO2-CO2 and CO2-TiO2 surface interactions in both of the magnitude of binding free energy and the interacting configuration.
Figure 8. The potential of mean force (PMF) in the range of the first peak from g(r) of (a) C-OC and (b) C-Ob for the coverages of 1/3, 1/2, 2/3 and 1 ML, respectively. 63 In Applications of Molecular Modeling to Challenges in Clean Energy; Fitzgerald, G., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2013.
Downloaded by UNIV OF MINNESOTA on October 14, 2014 | http://pubs.acs.org Publication Date (Web): June 3, 2013 | doi: 10.1021/bk-2013-1133.ch004
Conclusions The overall behavior of CO2 molecules on TiO2 surface can be summarized as follows. At first, OV defect sites are the most favorable adsorption sites for CO2 with the binding energy of 0.60 eV. Additional CO2 can be stabilized stronger next to OV bound CO2 on neighboring Ti5C site and then additional CO2 molecules start populating on regular Ti5C sites. When CO2 molecules are adsorbed with low coverage, the isolated CO2 molecule binds to one of the O atom on Ti5C site while it is tilted toward nearest neighboring Ob row with a binding energy of 0.45 eV. This adsorbed molecule can rotate with a low barrier of 0.05 eV via flat-lying configuration in between two neighboring Ti5C sites. At the same time, this intermediate configuration is able to participate in tumbling of CO2 to the next Ti5C site with a low barrier of 0.06 eV and further diffusion is possible in the same manner along the Ti5C row. As the coverage increases, CO2 molecule tends to be paired with another CO2 molecule while their configurations are tilted away from each other. At 1 ML coverage, the configuration of each CO2 molecule is governed by competition between attractive quadrupole-quadrupole interaction and steric repulsion. Note that the CO2 resided on Ob site is also available with being stabilized by neighboring CO2 molecules. As a result, the potential energy surface on TiO2 surface is corrugated in the presence of weak interactions characterized by CO2 interactions with another CO2 and TiO2 surface. In addition, relatively low barriers for CO2 rotation and diffusion can lead to a partial ordering of molecules at any instance. Extending each configuration to whole ensemble at the same time, the averaged features can show the regular pattern at the high coverage. In conclusion, the behavior of CO2 molecule on the TiO2 surface can be characterized by highly dynamic nature, and the fine structures are generated by the competition of interactions of CO2 with neighboring CO2 as well as with TiO2 surface. The present molecular modeling approach can provide useful information particularly when it is applied to various experimental observations to be collectively interpreted (7).
Acknowledgments The author would like to thank Dr. Roger Rousseau for the helpful discussions and advices on the present study and the preparation of this manuscript. This work is supported by the US Department of Energy (DOE), Office of Basic Science, Division of Chemical Sciences, Geosciences and Biosciences. Pacific Northwest National Laboratory (PNNL) is multiprogram national laboratory operated for DOE by Battelle. A portion of the research was performed using EMSL, a national scientific user facility sponsored by the Department of Energy’s Office of Biological and Environmental Research and located at Pacific Northwest National Laboratory.
References 1.
Chu, S.; Majumdar, A Nature 2012, 488, 294–303. 64 In Applications of Molecular Modeling to Challenges in Clean Energy; Fitzgerald, G., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2013.
2. 3. 4. 5.
Downloaded by UNIV OF MINNESOTA on October 14, 2014 | http://pubs.acs.org Publication Date (Web): June 3, 2013 | doi: 10.1021/bk-2013-1133.ch004
6.
7.
8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28.
Indrakanti, V. P.; Kubicki, J. D.; Schobert, H. H. Energy Environ. Sci. 2009, 2, 745–758. Roy, S. C.; Varghese, O. K.; Paulose, M.; Grimes, C. A. ACS Nano 2010, 4, 1259–1278. Freund, H. J.; Roberts, M. W. Surf. Sci. Rep. 1996, 25, 225–273. Basic Research Needs: Catalysis for Energy. http://www.pnl.gov/main/ publications/external/technical_reports/PNNL-17214.pdf. Marx, D.; Hutter, J.; Ab Initio Molecular Dynamics: Theory and Implementation. In Modern Methods and Algorithms of Quantum Chemistry; Grotendorst, J., Ed.; John von Neumann Institute for Computing Series: Jülich, 2000; Vol. 3, pp 329−477. Lin, X.; Yoon, Y.; Petrik, N. G.; Li, Z.; Wang, Z. T.; Glezakou, V. A.; Kay, B. D.; Lyubinetsky, I.; Kimmel, G. A.; Rousseau, R.; Dohnálek, Z. J. Phys. Chem. C 2012, DOI: 10.1021/jp308061j. Inoue, T.; Fujishima, A.; Konishi, S.; Honda, K. Nature 1979, 277, 637–638. Pang, C. L.; Lindsay, R.; Thornton, G. Chem. Soc. Rev. 2008, 37, 2328–2353. Thompson, T. L.; Diwald, O.; Yates, J. T. J. Phys. Chem. B 2003, 107, 11700–11704. Henderson, M. A. Surf. Sci. 1998, 400, 203–219. Lee, J.; Sorescu, D. C.; Deng, X.; Jordan, K. D. J. Phys. Chem. Lett. 2011, 2, 3114–3117. Lee, J.; Sorescu, D. C.; Deng, X. Y. J. Am. Chem. Soc. 2011, 133, 10066–10069. Tan, S. J.; Zhao, Y.; Zhao, J.; Wang, Z.; Ma, C. X.; Zhao, A. D.; Wang, B.; Luo, Y.; Yang, J. L.; Hou, J. G. Phys. Rev. B 2011, 84, 155418. Sorescu, D. C.; Lee, J.; Al-Saidi, W. A.; Jordan, K. D. J. Chem. Phys. 2011, 134, 104707. Acharya, D. P.; Camillone, N.; Sutter, P. J. Phys. Chem. C 2011, 115, 12095–12105. Dohnálek, Z.; Lyubinetsky, I.; Rousseau, R. Prog. Surf. Sci. 2010, 85, 161–205. Henderson, M. A. Surf. Sci. Rep. 2002, 46, 5–308. Diebold, U. Surf. Sci. Rep. 2003, 48, 53–229. The CP2K Developers Group, 2009. http://cp2k.org/. VandeVondele, J.; Krack, M.; Mohamed, F.; Parrinello, M.; Chassaing, T.; Hutter, J. Comput. Phys. Commun. 2005, 167, 103–128. Lippert, G.; Hutter, J.; Parrinello, M. Mol. Phys. 1997, 92, 477–487. Perdew, J. P.; Burke, K.; Ernzerhof, M. Phys. Rev. Lett. 1996, 77, 3865–3868. Goedecker, S.; Teter, M.; Hutter, J. Phys. Rev. B 1996, 54, 1703–1710. VandeVondele, J.; Hutter, J. J. Chem. Phys. 2007, 127, 114105. Grimme, S. J. Comp. Chem. 2006, 27, 1787–1799. Zhang, Z. R.; Rousseau, R.; Gong, J. L.; Kay, B. D.; Dohnálek, Z. J. Am. Chem. Soc. 2009, 131, 17926–17932. Kwak, J. H.; Rousseau, R.; Mei, D. H.; Peden, C. H. F.; Szanyi, J. ChemCatChem 2011, 3, 1557–1561. 65 In Applications of Molecular Modeling to Challenges in Clean Energy; Fitzgerald, G., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2013.
Downloaded by UNIV OF MINNESOTA on October 14, 2014 | http://pubs.acs.org Publication Date (Web): June 3, 2013 | doi: 10.1021/bk-2013-1133.ch004
29. Glezakou, V. A.; Rousseau, R.; Dang, L. X.; McGrail, B. P. Phys. Chem. Chem. Phys. 2010, 12, 8759–8771. 30. Windisch, C. F.; Glezakou, V. A.; Martin, P. F.; McGrail, B. P.; Schaef, H. T. Phys. Chem. Chem. Phys. 2012, 14, 2560–2566. 31. Henkelman, G; Uberuaga, B. P.; Jonsson, H. J. Chem. Phys. 2000, 113, 9901–9904. 32. Nosé, S. J. Chem. Phys. 1984, 81, 511–519. 33. Hoover, W. G. Phys. Rev. A 1985, 31, 1695–1697. 34. Dudarev, S. L.; Botton, G. A.; Savrasov, S. Y.; Humphreys, C. J.; Sutton, A. P. Phys. Rev. B 1998, 57, 1505–1509. 35. Borodin, A.; Reichling, M. Phys. Chem. Chem. Phys. 2011, 13, 15442–15447. 36. Yim, C. M.; Pang, C. L.; Thornton, G. Phys. Rev. Lett. 2010, 104, 036806. 37. Di Valentin, C.; Pacchioni, G.; Selloni, A. Phys. Rev. Lett. 2006, 97, 166803. 38. Deskins, A.; Rousseau, R.; Dupuis, M. J. Phys. Chem. C 2011, 115, 7562–7572. 39. Saharay, M; Balasubramanian, S. J. Chem. Phys. 2004, 120, 9694–9702. 40. Santoro, M.; Gorelli, F. A. Chem. Soc. Rev. 2006, 35, 918–931. 41. McQuarrie, D. A. Statistical Mechanics; University Science Books: Sausalito, CA, 2000; pp 261−264.
66 In Applications of Molecular Modeling to Challenges in Clean Energy; Fitzgerald, G., et al.; ACS Symposium Series; American Chemical Society: Washington, DC, 2013.